JP2002304288A - Data processing device and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a data processing device capable of enhancing, using a simple arrangement, the computational accuracy of data rounding-off that is based on block floating method. SOLUTION: Computations are made for each data of a data group, a minimal scale factor representative of these computation data is detected as a group scale factor(GSF), and scaling is carried out for each computation data based on the GSF detected. These processes are carried out for each data group of a data block, and the smallest of the GSFs detected is detected as a block scale factor (BSF). When computations are to be made again for the computation data, scaling is carried out for each computation data of the data group by a second shifter 18 according to the GSF and the BSF of the data group prior to the computations, and these computation data scaled are rounded off.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for forming a data group including one or a plurality of data, forming a data block including a plurality of data groups, and processing the data in data block units. More particularly, the present invention relates to a block-floating digital signal processor (hereinafter, simply referred to as DSP) that performs fixed-point arithmetic in a block-floating method.

[0002]

2. Description of the Related Art As a numerical expression method in digital signal processing, there are a fixed-point expression method and a floating-point expression method. In the floating point representation method, each data has an exponent part and a mantissa part, which has an advantage that high accuracy and a wide dynamic range can be secured, but has a problem that complicated and large-scale hardware is required. . On the other hand, in the fixed-point representation method, the hardware is simple and the circuit scale is small, but there is a problem in that the calculation accuracy is reduced. Therefore, a method called block floating has been proposed as a countermeasure for these problems. In this method, a predetermined number (for example, m) of data is regarded as one data block, a data block is provided with one block scale factor, and a common scaling (hereinafter, referred to as block) is applied to m data of the data block. (Referred to as normalization)) to effectively use a limited dynamic range and suppress deterioration in accuracy.

In order to efficiently realize block floating, for example, in an international application (International Publication No. WO99 / 66423) published based on the Patent Cooperation Treaty, a data operation processing device as shown in FIG. 8 is proposed. ing. FIG.
FIG. 10 is a block diagram illustrating a configuration of a conventional data processing device. As shown in FIG. 8, the data arithmetic processing device includes a data memory 100 for storing data in data block units, an arithmetic unit 200 for performing a multiply-accumulate operation on data in the data memory 100, and a block floating unit. A control unit 300 for performing necessary normalization, a control logic 400 for performing these controls, and an address generator 500
It is composed of Data memory 100, arithmetic unit 20
0, the control unit 300 and the control logic 400 are connected to each other via the data bus 110 so as to be able to exchange data. They are connected so that they can be exchanged.

The data memory 100 receives data from outside (not shown), and stores the input data in data block units as data to be operated by the operation unit 200 as shown in FIG. . FIG. 9 is a diagram illustrating a data structure of data serving as an operation unit in the data memory 100. In the data memory 100, FIG.
As shown in (1), data is stored as a data group in combination with a number (in this case, k) corresponding to the number of calculation units, and a plurality of data groups are stored as a data block in combination. That is, the data block is configured to include a plurality of data groups,
The data group is configured to include data of a number corresponding to the number of calculation units.

[0005] calculation unit 200 is constituted by the k calculation units P ₀ to P _k-1 having the same function. Each compute unit P ₀ ~P _k-1, the data of one data group is allocated respectively, each computing unit P ₀ ~P _k-1 is
Process data in parallel in data group units. Each of the calculation units P _{0 to} P _k−1 is a second shifter 1 that scales data based on a given scale factor.
0, a product-sum operation unit 12 that performs a product-sum operation on the data from the second shifter 10, and a unit scale factor detector 14 that calculates the scale factor of the operation data from the product-sum operation unit 12. And a first shifter 16 for scaling the operation data from the product-sum operation unit 12 based on the scale factor.

The unit scale factor detector 14 is
The number of redundant bits of the operation data from the product-sum operation unit 12 is calculated, and this is calculated as a scale factor. Here, the scale factor is a shift amount when data is bit-shifted, and indicates the number of bits to be shifted. Hereinafter, the same applies to the group scale factor and the block scale factor.

The second shifter 10 performs a bit shift (hereinafter simply referred to as right shift) on given data in the direction of lower bits with a shift amount corresponding to a given scale factor. I have. First shifter 16
Is a bit shift (hereinafter simply referred to as left shift) in the direction of the higher-order bits on the operation data from the product-sum operation unit 12 with a shift amount corresponding to a given scale factor. Also, the first shifter 16
Is stored in the data memory 100 in order to prepare for the next block processing. Since the number of bits of the data in the data memory 100 is smaller than that of the calculation data from the first shifter 16, the 1 shifter 16
Takes the number of bits of the data in the data memory 100 based on the most significant bit of the left-shifted operation data, truncates the lower bits, and rounds the result as the operation data of the number of bits of the data in the data memory 100. It is supposed to do.

The control unit 300 includes a group scale factor detector 50 for detecting a group scale factor from the scale factor calculated by the unit scale factor detector 14, a group scale factor register file 52 for storing the group scale factor, It comprises a block scale factor detector 54 for detecting a block scale factor from a factor, a block scale factor register 56 for storing a block scale factor, and a digit adder 60 for calculating a difference between a group scale factor and a block scale factor. Have been.

The group scale factor detector 50 includes:
Among the scale factors calculated by the unit scale factor detectors 14 of the calculation units P _{0 to} P _k−1 , the smallest one is detected as the group scale factor, and the detected group scale factor is controlled by the control logic 400. Register file 52
And it is adapted to output to the first shifter 16 for each computing unit P ₀ ~P _k-1. For example, each calculation unit P ₀
ＰP _{k -1} (for example, k = 3) from the unit scale factor detector 14, "1" as a scale factor,
When “2” and “3” are input, the minimum value “1” is detected as the group scale factor.

The group scale factor register file 52 temporarily stores the group scale factors detected by the group scale factor detector 50, and the data memory 1 is controlled by the control logic 400.
The group scale factor is output to 00 or the digit adder 60. The block scale factor detector 54 detects the smallest one of the group scale factors detected by the group scale factor detector 50 as a block scale factor. For example, when the group scale factor of a certain data block is detected as “3”, “4”, or “5”, the minimum value “3” is detected as the block scale factor.

Block scale factor register 56
Is the block detected by the block scale factor detector 54.
The lock scale factor is temporarily stored and the control logic
Control of the data memory 100 or the digit
Output the group scale factor to the adder 60
It has become so. The digit matching adder 60 includes a glue.
Output from the scale factor register file 52.
Group scale factor and block scale factor
Block scale file output from the
And the block size from the group scale factor.
Calculate the difference by subtracting the scale factor and control
Under the control of the logic 400, the scale
Is calculated for each calculation unit P ₀~ P_k-1Second shifter 1
0 is output.

The control logic 400 includes the data memory 10
When an operation is performed on data of 0 for the first time, the following control processing is executed. First, data is read from the data memory 100 in data block units, and for each data group of the data block, the data of the data group is transferred to the second shifter 10 of each of the calculation units P _{0 to} P _k−1 via the data bus 110. Distribute.

And a group scale factor detector.
The group scale factor detected in step 50 is calculated for each calculation unit.
P₀~ P_k-1Of each calculation unit.
To P ₀~ P_k-1Calculation data from the first shifter 16
The data is stored in the data memory 100 via the tabus 110. Ma
In addition, the control logic 400 stores data in the data memory 100.
When the calculation is performed again on the data, the following control processing is executed.
It is supposed to.

[0014] First, read the operation data in a data block from the data memory 100, for each data group of data blocks, each computing units P ₀ ~ operation data of the data group through the data bus 110, respectively
P _k−1 is given to the second shifter 10. At the same time, the group scale factor and the block scale factor of the data supplied to the second shifters 10 of the respective calculation units P _{0 to} P _k−1 are stored in the group scale factor register file 5.
2 and stored in the block scale factor register 56, the block scale factor is read out from the block scale factor register 56 and supplied to the adder 60 for digit alignment, and the group scale factor is sequentially read out from the group scale factor register file 52. given the digit alignment adder 60, the difference between the group scale factor and the block scale factor calculated by alignment shift adder 60, a scale factor consisting of the difference of each calculation unit P ₀ ~P _k-1 2 to the shifter 10.

And a group scale factor detector.
The group scale factor detected in step 50 is calculated for each calculation unit.
P₀~ P_k-1Of each calculation unit.
To P ₀~ P_k-1Calculation data from the first shifter 16
The data is stored in the data memory 100 via the tabus 110.

[0016]

As described above, in order to apply the data rounding process in the block floating method, since the data rounding process is performed on the calculation result, the conventional data calculation process described above is performed. As in the case of a device, a configuration in which data rounding processing is performed on operation data shifted by the first shifter 16 is conceivable.

However, in the above-mentioned conventional data processing device, since the data rounding process is performed on the calculation data shifted by the first shifter 16, each calculation data has its decimal point. Data rounding processing is performed at different digit positions based on the position. For example, when the 16-bit operation data whose decimal point position is between the upper 2 bits and the 3 bits is shifted left by 3 bits by the first shifter 16, and when shifted by 2 bits, the data rounding process is performed. The digit position to be shifted is shifted by one digit with respect to the decimal point position. Therefore, it has not been possible to sufficiently improve the calculation accuracy by rounding the data.

On the other hand, a configuration in which the data is rounded before shifting by the first shifter 16 is also conceivable.
Since the digit position at which the data is rounded varies depending on the value of the group scale factor and is not constant, a complicated configuration is required to realize this. Therefore, the present invention has been made by focusing on such unresolved problems of the conventional technology, and aims to improve calculation accuracy by rounding data with a simple configuration in a block floating system. It is an object of the present invention to provide a data operation processing device and a data operation processing program which can perform the operation.

[0019]

According to a first aspect of the present invention, there is provided a data processing apparatus comprising: a data group including one or a plurality of data; A data processing apparatus comprising a data block including a group and processing the data in units of the data block, performing an operation on each data of the data group, and a scale of the operation data having a maximum absolute value. Performing a series of these processes of detecting the factor as a group scale factor and performing scaling on each of the operation data based on the detected group scale factor for each data group of the data block; Block the factor corresponding to the operation data with the largest absolute value When the data is detected as a scale factor and the calculation is performed again on the scaled calculation data, before the calculation, each calculation data of the data group is subjected to the group scale factor of the data group and the block scale factor. The scaling is performed on the basis of the calculated data, and the scaled data is rounded.

With such a configuration, the arithmetic operation is performed on each data of the data group, and the scale factor of the arithmetic data having the maximum absolute value is detected as the group scale factor, and the detected group is determined. Each operation data is scaled based on the scale factor. Such a series of processing is performed for each data group of the data block. Then, when the processing is completed for one data block,
Among the group scale factors detected for each data group, the one corresponding to the operation data having the maximum absolute value is detected as the block scale factor.

Next, when the operation is performed again on the scaled operation data, each operation data of the data group is scaled based on the group scale factor and the block scale factor of the data group. Data rounding processing is performed on the scaled operation data. In this data rounding process, the operation data belonging to the same data group has the same group scale factor and block scale factor, and therefore, the digit position for performing the data rounding process is the same by scaling based on them. .

Here, as the scale factor, for example, a shift amount when bit-shifting data can be cited. In this case, scaling is performed by performing a bit shift on the data with a shift amount corresponding to the scale factor. Hereinafter, the same applies to the data arithmetic processing device according to claim 6. As a configuration for detecting the group scale factor, for example, after calculating each data of the data group, the scale factor of the calculated data is calculated, and the smallest one of the calculated scale factors is used as the group scale factor. The detection may be performed as a factor, or the scale factor of the operation data may not be calculated, and the group scale factor may be directly detected from the operation data.

Furthermore, a data operation processing device according to a second aspect of the present invention is the data operation processing device according to the first aspect, wherein the operation is performed on data having a predetermined number of bits. For each operation data scaled based on the group scale factor and the block scale factor, from the operation data,
Extracting data having a bit number larger than the predetermined number of bits based on the most significant bit, and extracting the data based on values of data other than the predetermined number of bits based on the most significant bit of the extracted data. Rounding processing is performed.

With this configuration, when the operation is performed again on the scaled operation data, each operation data of the data group is calculated based on the group scale factor and the block scale factor of the data group. Scaling is performed. Then, for each of the scaled operation data, data having a bit number greater than a predetermined number of bits is extracted from the operation data with reference to the most significant bit, and based on the most significant bit of the extracted data. Data rounding processing is performed based on values other than data having a predetermined number of bits.

Further, according to a third aspect of the present invention, in the data operation processing device according to the second aspect, for each operation data scaled based on the group scale factor and the block scale factor. Extracting, from the operation data, data having a bit number that is one bit larger than the predetermined bit number based on the most significant bit, and extracting data other than the predetermined bit number data based on the most significant bit from the extracted data. The data is rounded based on the value of the object.

With this configuration, when the operation is performed again on the scaled operation data, each operation data of the data group is calculated based on the group scale factor and the block scale factor of the data group. Scaling is performed. Then, for each of the scaled operation data, data having a bit number that is one bit larger than a predetermined bit number is extracted from the operation data with reference to the most significant bit, and the most significant bit of the extracted data is extracted. Data rounding is performed based on values other than data having a predetermined number of bits as a reference.

Further, according to a fourth aspect of the present invention, in the data arithmetic processing apparatus according to the third aspect, the rounding of the data is performed based on the most significant bit of the extracted data. , One bit is added to the value of data other than the data of the predetermined number of bits, and the data of the predetermined number of bits is output as an operation target based on the most significant bit of the addition result.

With such a configuration, in the data rounding process, one bit is added to the value of data other than data having a predetermined number of bits based on the most significant bit of the extracted data, and the result of the addition is obtained. Among them, data of a predetermined number of bits is output as a calculation target based on the most significant bit. Further, in the data processing device according to claim 5 of the present invention, in the data processing device according to any one of claims 1 to 4, when performing the calculation again on the scaled calculation data, Prior to the operation, scaling is performed on each operation data of the data group based on a difference between a group scale factor of the data group and the block scale factor.

With this configuration, when the operation is performed again on the scaled operation data, the difference between the group scale factor of the data group and the block scale factor is calculated for each operation data of the data group. The scaling is performed based on Furthermore, a data processing device according to claim 6 of the present invention includes a plurality of calculation units, configures a data group including one or a plurality of data, and configures a data block including a plurality of data groups. And an apparatus for processing data on a data block basis, wherein each of the calculation units performs scaling on data based on a given scale factor; and Calculating means for performing an operation on the data, scale factor calculating means for calculating a scale factor of the calculated data from the calculating means, and calculating data from the calculating means based on another given scale factor. Second scaling means for performing scaling, and further comprising a memory for storing data. Means, a group scale factor detecting means for detecting, as a group scale factor, a scale factor calculated by the scale factor calculating means of each of the calculation units, the one corresponding to the operation data having the largest absolute value, and the group scale factor detecting means A block scale factor detecting means for detecting, as a block scale factor, the one corresponding to the operation data having the absolute value of the group scale factor detected by the means, and a control means for performing control for data processing, The control means reads data from the storage means in units of the data block, distributes the data of the data group to the first scaling means of each of the calculation units for each data group of the data block, and The group scale factor detected by the factor detection means is given to the second scaling means of each of the calculation units, the operation data from the second scaling means of each of the calculation units is stored in the storage means, and the operation data of the storage means is stored in the storage means. When the calculation is performed again, for each data group of the data block, the calculation data of the data group is distributed to the first scaling means of each of the calculation units, and the group scale factor of the data group and the block A scale factor consisting of a difference from a scale factor is provided to first scaling means of each of the calculation units, and the first scaling means performs scaling on the data based on the given scale factor. hand The data is rounded.

With such a configuration, the control means reads data from the storage means in data block units, and for each data group of the data block, the data of the data group is firstly scaled by each calculation unit. Allocated to means. In each of the calculation units, the scaling factor is not given to the first scaling means in the initial state, so that the scaling is not performed by the first scaling means and is directly input to the calculation means. Then, the operation unit performs an operation on the data from the first scaling unit, and the scale factor calculation unit calculates the scale factor of the operation data from the operation unit.

When the scale factor is calculated by each calculation unit, the group scale factor detecting means
Of the scale factors calculated by the scale factor calculation means of each calculation unit, the one corresponding to the operation data having the maximum absolute value is detected as a group scale factor, and the control means converts the calculated group scale factor into each calculation unit. To the second scaling means.

Thus, in each calculation unit, the second scaling means scales the operation data from the operation means based on the given group scale factor. As described above, when scaling is performed on the operation data in each calculation unit, the control unit stores the operation data from the second scaling unit of each calculation unit in the storage unit, and the block scale factor detection unit Among the group scale factors detected by the group scale factor detection means, the one corresponding to the operation data having the maximum absolute value is detected as the block scale factor.

Next, when the arithmetic operation is performed again on the arithmetic data in the storage means, the arithmetic means reads out the arithmetic data from the storage means in data block units by the control means. Is distributed to the first scaling means of each calculation unit, and a scale factor comprising a difference between the group scale factor of the data group and the block scale factor is given to the first scaling means of each calculation unit.

In each calculation unit, the given operation data is scaled by the first scaling means on the basis of the scale factor comprising the difference between the group scale factor and the block scale factor detected in the previous processing. Then, data rounding processing is performed on the scaled operation data. In this data rounding process, the operation data belonging to the same data group has the same group scale factor and block scale factor, and therefore, the digit position for performing the data rounding process is the same by scaling based on them. .

Here, the storage means stores data by all means and at all times, and may store data in advance, or may store the data without storing the data in advance. During the operation, data may be stored by an external input or the like. The data processing unit according to claim 7 of the present invention is the data processing unit according to claim 6, wherein the operation means performs an operation on data of a predetermined number of bits, and the first scaling means Extracts, from the scaled data, data having a larger number of bits than the predetermined number of bits on the basis of the most significant bit, and among the extracted data, data other than the data of the predetermined number of bits on the basis of the most significant bit. The data is rounded based on the value of the object.

With such a configuration, when the arithmetic operation is performed again on the arithmetic data in the storage means, the arithmetic means reads out the arithmetic data in units of data blocks from the storage means by the control means, and the data groups of the data blocks are read out. Every
The operation data of the data group is respectively distributed to the first scaling means of each calculation unit, and a scale factor comprising a difference between the group scale factor and the block scale factor of the data group is provided to the first scaling means of each calculation unit. .

In each calculation unit, the given operation data is scaled by the first scaling means on the basis of the scale factor consisting of the difference between the group scale factor and the block scale factor detected in the previous processing. Is Then, from the scaled operation data, data having a larger number of bits than the predetermined number of bits is extracted based on the most significant bit, and other than the predetermined number of bits of data of the extracted data based on the most significant bit. Data rounding is performed based on the value of the object.

Further, in the data processing apparatus according to claim 8 of the present invention, in the data processing apparatus according to claim 7, the first scaling means uses the most significant bit as a reference from the scaled data. Extracting data having a bit number one bit larger than the predetermined number of bits, and performing a rounding process of the data based on a value of data other than the predetermined number of bits based on the most significant bit of the extracted data. It has become.

With this configuration, when the operation is performed again on the operation data in the storage unit, the operation unit reads out the operation data in units of data blocks from the storage unit, and reads each data group of the data block. Every
The operation data of the data group is respectively distributed to the first scaling means of each calculation unit, and a scale factor comprising a difference between the group scale factor and the block scale factor of the data group is provided to the first scaling means of each calculation unit. .

In each of the calculation units, the given operation data is scaled by the first scaling means on the basis of the scale factor comprising the difference between the group scale factor and the block scale factor detected in the previous processing. Is Then, from the scaled operation data, data having a bit number one bit larger than the predetermined bit number is extracted based on the most significant bit, and the extracted data having a predetermined bit number based on the most significant bit is extracted. Data rounding is performed based on values other than the above.

Further, according to a ninth aspect of the present invention, in the data arithmetic processing apparatus according to the eighth aspect, the rounding of the data is performed based on the most significant bit of the extracted data. , One bit is added to the value of data other than the data of the predetermined number of bits, and the data of the predetermined number of bits is output as an operation target based on the most significant bit of the addition result.

With such a configuration, in the data rounding process, one bit is added to the value of data other than data having a predetermined number of bits based on the most significant bit of the extracted data, and the result of the addition is obtained. Among them, data of a predetermined number of bits is output as a calculation target based on the most significant bit. Further, according to a tenth aspect of the present invention, in the data operation processing device according to any one of the sixth to ninth aspects, a second data processing device for storing the group scale factor and the block scale factor. A storage unit, wherein the control unit stores the group scale factor and the block scale factor in the second storage unit in association with operation data, and performs the operation again on the operation data in the storage unit. Reading the corresponding group scale factor and block scale factor from the second storage means, and for each data group of the data block, allocating the operation data of the data group to the first scaling means of each of the calculation units, Group scale of data group The scale factor consisting of the difference between said block scale factor Akuta adapted to provide the first scaling means of each calculation unit.

With such a configuration, the control means converts the group scale factor detected by the group scale factor detection means and the block scale factor detected by the block scale factor detection means into:
The data is stored in the second storage unit in association with the operation data from the second scaling unit of each calculation unit. When the calculation is again performed on the calculation data in the storage means, the corresponding group scale factor and block scale factor are read from the second storage means by the control means, and the data is stored in each data group of the data block. The operation data of the group is respectively distributed to the first scaling means of each calculation unit, and a scale factor comprising a difference between the group scale factor and the block scale factor of the data group is provided to the first scaling means of each calculation unit.

Thus, in each of the calculation units, the first scaling means performs processing on the given operation data based on the scale factor comprising the difference between the group scale factor and the block scale factor detected in the previous processing. Scaling is performed. Here, the second storage means stores the group scale factor and the block scale factor by any means and at any time, and may store the group scale factor and the block scale factor in advance. Instead of storing the group scale factor and the block scale factor in advance, the group scale factor and the block scale factor may be stored by an external input or the like during the operation of the present apparatus.

Further, according to the present invention, there is provided a data processing apparatus according to any one of claims 6 to 10, wherein the scale factor calculating means is configured to perform a calculation from the calculating means. The number of redundant bits of data is calculated, and this is calculated as a scale factor. With such a configuration, the scale factor calculation means calculates the number of redundant bits of the operation data from the operation means, and this is calculated as the scale factor.

Further, according to a twelfth aspect of the present invention, in the data arithmetic processing device according to the eleventh aspect, the arithmetic means is a fixed-point arithmetic unit, and the first scaling means is A bit shift is performed on the data by a shift amount corresponding to a given scale factor, and the second scaling means converts the data from the arithmetic means into a shift data corresponding to another given scale factor. A bit shift is performed for this.

With such a configuration, in each of the calculation units, the first scaling means performs a bit shift on the given data with a shift amount corresponding to the given scale factor, and , Fixed-point arithmetic is performed on the data from the first scaling means. Then, the second scaling means performs a bit shift on the operation data from the operation means by a shift amount corresponding to the given scale factor.

Here, the first scaling means shifts the data to the right, for example.
Further, the second scaling means performs, for example, a left shift or a right shift on the data. According to a thirteenth aspect of the present invention, in the data arithmetic processing device according to any one of the sixth to twelfth aspects, the arithmetic means includes: This is a product-sum calculator for calculating the product sum with a coefficient.

With such a configuration, the calculating means calculates the product sum of the data from the first scaling means and the predetermined coefficient, and outputs the product sum data as the calculated data. On the other hand, to achieve the above object, a data operation processing program according to claim 14 of the present invention forms a data group including one or a plurality of data, and further forms a data block including a plurality of data groups. A computer-executable program configured to process data in units of the data blocks, performing an operation on each data of the data group, and determining a scale factor of an operation data having a maximum absolute value in the group. A series of these processes for detecting as a scale factor and performing scaling on the respective calculation data based on the detected group scale factor,
Performed for each data group of the data block, and further detects, as the block scale factor, the detected group scale factor corresponding to the calculated data having the largest absolute value, and performs the calculation again on the scaled calculated data. Before performing the calculation, scaling is performed on each calculation data of the data group based on the group scale factor and the block scale factor of the data group, and data of the scaled calculation data is Rounding processing is performed.

With such a configuration, when the program is read by the computer and the computer executes processing in accordance with the read program,
An operation equivalent to that of the described data processing device can be obtained. Further, a data operation processing program according to claim 15 of the present invention provides a computer having storage means for storing data to form a data group including one or a plurality of data, A first scaling unit configured to form a data block including a group and to process data in units of the data block, wherein the first scaling unit performs scaling on the data based on a given scale factor; Calculating means for calculating the scale factor of the calculation data from the calculation means, and calculation data from the calculation means based on another given scale factor. Implemented as a second scaling means for performing scaling A program for executing the processing for each of the data groups, and a scale factor calculated by the scale factor calculating means corresponding to each of the data groups, the one corresponding to the operation data having the largest absolute value is defined as a group scale factor. Group scale factor detecting means for detecting, block scale factor detecting means for detecting, as a block scale factor, a group scale factor detected by the group scale factor detecting means corresponding to the operation data having the maximum absolute value, and data processing And a program for executing processing realized as control means for performing control for the data block. The control means reads data from the storage means in units of the data block, and reads each data group of the data block. Each time, the data of the data group is distributed to the first scaling means corresponding to each data group, and the group scale factor detected by the group scale factor detection means is given to the second scaling means corresponding to each data group. When the calculation data from the second scaling means corresponding to each data group is stored in the storage means, and the calculation is again performed on the calculation data in the storage means, for each data group of the data block, The operation data of the data group is distributed to the first scaling means corresponding to each data group, and a scale factor including a difference between a group scale factor of the data group and the block scale factor is assigned to each data group. The data is provided to a corresponding first scaling means, and the first scaling means performs scaling on the data based on the given scale factor, and performs a rounding process on the scaled data.

With such a configuration, when the program is read by the computer, and the computer executes the processing in accordance with the read program, the present invention can be applied to any of the above embodiments.
An operation equivalent to that of the described data processing device can be obtained. The outline of the present invention is, for example, as shown in FIG. In the present invention, one scale factor is calculated for each data group, and this is used as a group scale factor, and the data output from each of the product-sum calculators 12 is normalized by the group scale factor (hereinafter, referred to as group normalization). And
The group scale factor is stored in the register file 52 in association with the group-normalized data group. Such a series of processing is repeatedly applied to other data groups included in the data block. After processing one data block, the smallest scale factor among the plurality of group scale factors is detected and used as a block scale factor, which is stored in the register file 56 in association with the data block. Thereafter, when performing block processing on this data block, since the block scale factor and the group scale factor are different and the digit positions may not be aligned between the data groups, the difference between the group scale factor and the block scale factor is used. Then, the data of each data group is normalized (hereinafter, referred to as block normalization) to perform digit matching, and after performing the data rounding process for each of the data subjected to the digit matching, the product-sum calculator 12 input.

As a result, the data after the end of the product-sum operation is
Since the data is once group-normalized in units of data groups and stored in the data memory 100, and used in the next block processing, the data is subjected to block normalization. It is not included, and the calculation error when performing the fixed-point calculation by the block floating method is reduced. Further, it is possible to improve the calculation accuracy by rounding the data.

[0053]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 to FIG. 5 are views showing an embodiment of a data operation processing device according to the present invention.
In this embodiment, a data operation processing device and a data operation processing program according to the present invention are applied to a case where a fixed-point operation is performed by a block floating method as shown in FIG. The difference from the processing device is that data is rounded by the second shifter 18 instead of the first shifter 16.

First, the configuration of the data processing unit according to the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing the configuration of the data processing device according to the present invention. As shown in FIG. 1, the data arithmetic processing device according to the present invention includes a data memory 100 for storing data in data block units, an arithmetic unit 200 for performing a product-sum operation on data in the data memory 100, The control unit 300 includes a control unit 300 for performing normalization required for block floating, a control logic 400 for performing these controls, and an address generator 500. Data memory 100, arithmetic unit 200, control unit 300, and control logic 400
Are connected to each other via a data bus 110 so that data can be exchanged. The data memory 100, the control unit 300, and the address generator 500
And are connected so as to be able to exchange data.

The data memory 100 receives data from the outside (not shown) and stores the input data in data block units as data to be operated by the operation unit 200 as shown in FIG. . In the data memory 100, as shown in FIG. 9, data is stored as a data group by combining a number (in this case, k) corresponding to the number of calculation units, and a plurality of data groups are combined to form a data block. Is stored. That is, the data block is configured to include a plurality of data groups, and the data group is configured to include data of a number corresponding to the number of calculation units.

The operation unit 200 is composed of k calculation units P _{0 to} P _k−1 having the same function. Each compute unit P ₀ ~P _k-1, the data of one data group is allocated respectively, each computing unit P ₀ ~P _k-1 is
Process data in parallel in data group units. Each of the calculation units P _{0 to} P _k−1 is a second shifter 1 that scales data based on a given scale factor.
8, a product-sum operation unit 12 for performing a product-sum operation on data from the second shifter 18, and a unit scale factor detector 14 for calculating a scale factor of the operation data from the product-sum operation unit 12. And a first shifter 16 for scaling the operation data from the product-sum operation unit 12 based on the scale factor.

The unit scale factor detector 14
The number of redundant bits of the operation data from the product-sum operation unit 12 is calculated, and this is calculated as a scale factor. Here, the scale factor is a shift amount when data is bit-shifted, and indicates the number of bits to be shifted. Hereinafter, the same applies to the group scale factor and the block scale factor.

The second shifter 18 shifts the given data to the right by a shift amount corresponding to the given scale factor. First shifter 16
Is a shift amount corresponding to a given scale factor, and shifts the operation data from the product-sum operation unit 12 to the left. Further, the operation data from the first shifter 16 is used in preparation for the next block processing.
Although stored in the data memory 100, the data memory 10
Since the number of bits of 0 data is smaller than that of the operation data from the first shifter 16, the first shifter 16 determines the number of bits of the data in the data memory 100 based on the most significant bit of the left-shifted operation data. Minutes, the lower bits are simply truncated, and the data memory 10
Operation data of the number of bits of 0 data is output.

The control unit 300 includes a unit scale factor.
From the scale factor calculated by the detector 14
Group scale factor to detect scale factor
A detector 50 and a group for storing a group scale factor.
Loop scale factor register file 52 and group
From block scale factor to block scale factor
A block scale factor detector 54 for detecting
Block scale factor that stores the
Register register 56, group scale factor and
Addition for digit alignment to calculate the difference of the clock scale factor
And a vessel 60.

The group scale factor detector 50 is
Among the scale factors calculated by the unit scale factor detectors 14 of the calculation units P _{0 to} P _k−1 , the smallest one is detected as the group scale factor, and the detected group scale factor is controlled by the control logic 400. Register file 52
And it is adapted to output to the first shifter 16 for each computing unit P ₀ ~P _k-1. For example, each calculation unit P ₀
ＰP _{k -1} (for example, k = 3) from the unit scale factor detector 14, "1" as a scale factor,
When “2” and “3” are input, the minimum value “1” is detected as the group scale factor.

The group scale factor register file 52 temporarily stores the group scale factors detected by the group scale factor detector 50, and the data memory 1 is controlled by the control logic 400.
The group scale factor is output to 00 or the digit adder 60. The block scale factor detector 54 detects the smallest one of the group scale factors detected by the group scale factor detector 50 as a block scale factor. For example, when the group scale factor of a certain data block is detected as “3”, “4”, or “5”, the minimum value “3” is detected as the block scale factor.

Block scale factor register 56
Is the block detected by the block scale factor detector 54.
The lock scale factor is temporarily stored and the control logic
Control of the data memory 100 or the digit
Output the group scale factor to the adder 60
It has become so. The digit matching adder 60 includes a glue.
Output from the scale factor register file 52.
Group scale factor and block scale factor
Block scale file output from the
And the block size from the group scale factor.
Calculate the difference by subtracting the scale factor and control
Under the control of the logic 400, the scale
Is calculated for each calculation unit P ₀~ P_k-1Second shifter 1
8 is output.

The control logic 400 includes the data memory 10
When an operation is performed on data of 0 for the first time, the following control processing is executed. First, data is read from the data memory 100 in data block units, and for each data group of the data block, the data of the data group is transferred to the second shifter 18 of each of the calculation units P _{0 to} P _k−1 via the data bus 110. Distribute.

Then, a group scale factor detector
The group scale factor detected in step 50 is calculated for each calculation unit.
P₀~ P_k-1Of each calculation unit.
To P ₀~ P_k-1Calculation data from the first shifter 16
The data is stored in the data memory 100 via the tabus 110. Ma
In addition, the control logic 400 stores data in the data memory 100.
When the calculation is performed again on the data, the following control processing is executed.
It is supposed to.

[0065] First, read the operation data in a data block from the data memory 100, for each data group of data blocks, each computing units P ₀ ~ operation data of the data group through the data bus 110, respectively
P _k−1 is given to the second shifter 18. At the same time, the group scale factor and the block scale factor of the data supplied to the second shifters 18 of the calculation units P _{0 to} P _k−1 are stored in the group scale factor register file 5.
2 and stored in the block scale factor register 56, the block scale factor is read out from the block scale factor register 56 and supplied to the digitizing adder 60, and the group scale factor is sequentially read out from the group scale factor register file 52. given the digit alignment adder 60, the difference between the group scale factor and the block scale factor calculated by alignment shift adder 60, a scale factor consisting of the difference of each calculation unit P ₀ ~P _k-1 2 shifter 18.

Then, the group scale factor detector
The group scale factor detected in step 50 is calculated for each calculation unit.
P₀~ P_k-1Of each calculation unit.
To P ₀~ P_k-1Calculation data from the first shifter 16
The data is stored in the data memory 100 via the tabus 110. What
Contact, each calculation unit P₀~ P_k-1To the second shifter 18
Group scale factor and block
Kale factor is the group scale factor register
File 52 and block scale factor register 5
6 is not stored in each calculation unit P₀~ P
_k-1Glue corresponding to the data given to the second shifter 18 of FIG.
Scale factor and block scale factor
Read from the data memory 100 and read them from the data bus
Group scale factor register file via 110
To block 52 and block scale factor register 56
After storing, the above processing is executed.

Next, the configuration of the second shifter 18 will be described in detail with reference to FIG. FIG. 2 is a block diagram showing a configuration of the second shifter 18. As shown in FIG. As shown in FIG. 2, the second shifter 18 shifts data right by a shift amount corresponding to a given scale factor.
And a data rounding section 18b for rounding the data from the right shifter 18a.

The right shifter 18a receives n-bit data from the data memory 100, shifts the input data to the right by a shift amount corresponding to the scale factor from the digit matching adder 60, and shifts the data. From the data, n + 1-bit data is extracted based on the most significant bit and output to the data rounding processing unit 18b. For example, when 8-bit data is shifted right by 3 bits, the data becomes 11-bit data, so that the upper 9 bits of the 11-bit data are output to the data rounding processing unit 18b.

The data rounding section 18b is provided with the right shifter 18
Input n + 1 bit data from a, and input n + 1
Addition data for data rounding processing in which only the least significant bit of the n + 1-bit data is set to "1" is added to the bit data, and n-bit data is extracted from the addition result based on the most significant bit. Output to the product-sum operation unit 12. For example, "01010101
When 9-bit data of "1" is input, the addition data "000000001" for data rounding processing is added to the input data, and the product-sum operation unit 12 outputs the upper 8 bits of "010101100" as the addition result. Will be output.

Next, the operation of the above embodiment will be described with reference to FIGS. In this example, the product-sum operation unit 1
In the case of No. 2, the first calculation is performed by the following equation (1), and the second and subsequent calculations on the result calculated by the following equation (1) are performed by the following equation (2). In the following equation (1), A (k) is coefficient data, and X (nk)
Is data from the second shifter 18. In the following equation (2), B (l) is coefficient data, and Y (m−1)
Is a result calculated by the following equation (1) and is data from the second shifter 18.

[0071]

(Equation 1)

[0072]

(Equation 2)

The data from the data memory 100 is 8 bits, and the output of the product-sum operation unit 12 is 16 bits.
Further, it is assumed that the arithmetic unit 200 is configured with four calculation units, the data block is configured to include two data groups, and the data group is configured to include four data corresponding to the number of calculation units. It is assumed that In addition, predetermined coefficient data A
Assume that (0) -A (3), B (0), B (1) are available whenever needed. How to supply these coefficient data is not directly described because it is not directly the essence of the present invention. Can be given.

First, a procedure for parallel processing of data by the following equation (1) in the data processing device according to the present invention will be described with reference to FIG. FIG. 3 shows each product-sum operation unit 1
FIG. 3 is a diagram showing a product-sum operation procedure performed in Step 2. Equation (1) above
When the operation is performed on the data in the data memory 100, as shown in FIG. 3, first, in a cycle 1, the first data group including the data X (0) to X (3) is Read, and data X (0)-
X (3) are distributed in the second shifter 18 of each calculation unit P ₀ to P ₃ through the data bus 110, respectively. In the initial state, since there is no scale factor corresponding to the data X (0) to X (3), the control logic 400 controls the second shifter 18 so that the data X (0) to X (3) is not shifted. Is done.

In each product-sum operation unit 12, coefficient data A (0) to A (3) and data X (0) to
The product-sum operation with X (3) is performed, and through cycle 2 to cycle 5, operation data Y (0) to Y (3), which are the first product-sum operation results, are obtained. In these arithmetic processes, the data X (0) to X (3) are stored in each calculation unit P ₀
Between to P ₃ shall be transmitted through hardware means.

The operation data Y (0) to Y (3) are thereafter input to each unit scale factor detector 14. In each unit scale factor detector 14, the number of redundant bits of the operation data from the product-sum operation unit 12 is calculated, and this is calculated as a scale factor. For example, in data “000001100”, since four “0” s continue from the most significant bit, the number of redundant bits is “4” and the scale factor is “4”.

The plurality of scale factors calculated by each unit scale factor detector 14 are then input to a group scale factor detector 50, where the smallest one of the plurality of scale factors is detected as a group scale factor. Is done. This group scale factor is further determined by the first shifter 1 of each of the calculation units P _{0 to} P ₃ .
6 is input. At the same time, the operation data Y (0) to Y (3) from each product-sum operation unit 12 are also input to the first shifter 16. According to the group scale factor, 16-bit operation data Y (0) to Y by the first shifter 16
Group normalization is performed on (3). FIG. 4 shows an example of a calculation result using actual numerical values.

These group-normalized operation data are stored in the data memory 100 in preparation for the next block processing. Since the number of bits of the data in the data memory 100 is smaller than that of the operation data, the lower 8 bits of each operation data are simply truncated. The operation data Y (0) to Y (3) from the first shifter 16 are shown in the input column of the data memory 100 in FIG. The operation data Y (0) to Y (3) are stored in the data memory 10
The group scale factor is also input to the group scale factor register file 52 during the storage to zero. This group scale factor is stored in the group scale factor register file 52 in association with the corresponding data group. For example, when a data group is stored in the data memory 100, a certain memory address is taken, and all data of the data group are simultaneously accessed by one memory address, and the operation data is stored in the data memory 100. In the register specified by the same address used to perform the operation. Alternatively, as shown in FIG. 5, a correspondence table between the address of the data memory 100 storing the data group and the address of the group scale factor register file 52 storing the group scale factor is stored in the data memory 100. It is also possible to create these sections and manage these addresses.

The group scale factor is also input to a block scale factor detector 54, where the smallest one of the plurality of group scale factors is detected as a block scale factor. When the calculation of the first data group of a certain data block is completed, the group scale factor just calculated, that is, “2” in the example of FIG.
It is stored in a register (not shown in FIG. 1) in the block scale factor detector 54.

Next, in cycle 6, data X (4) -X
The second data group including (7) is the data memory 1
Read from 00 is distributed to the second shifter 18 of each calculation unit P ₀ to P ₃ data X (4) ~X (7) is via the data bus 110, respectively. Data X (0) to X
The same operation as that for the first data group including (3) is performed for this data group. The only difference is that the group scale factor calculated for the operation data Y (4) to Y (7) is compared with the content held in the register in the block scale factor detector 54 this time, and Is selected as the block scale factor. In the example of FIG. 4, the group scale factor “1” is detected from the operation data Y (4) to Y (7). In this case, since “1” is smaller than “2” held in the register in the block scale factor detector 54, the contents of the register in the block scale factor detector 54 are updated.

In this example, since the operation on all the data of one data block has been completed at this point, the block scale factor is changed to the current value of the register in the block scale factor detector 54, that is, “1”. Is determined. As can be easily inferred from the present example, even if there are a plurality of data groups to be operated after this, the above-described processing applied to the second data group can be repeated.

Next, a procedure for parallel processing of the operation data calculated by the above equation (1) again by the following equation (2) in the data processing apparatus according to the present invention will be described. Before starting the operation, the contents of the registers in the block scale factor detector 54 are stored in the block scale factor register 5.
6 According to the above equation (2), the data memory 10
When the operation is performed again on the operation data 0, the first data group including the operation data Y (0) to Y (3) is read from the data memory 100, and the operation data Y
(0) to Y (3) are distributed to the second shifters 18 of the calculation units P _{0 to} P ₃ via the data bus 110, respectively. At the same time, the corresponding group scale factor, ie, “2” in this example, is selected by the address information provided via the address bus 120,
It is read from the group scale factor register file 52. The right shift amount in each second shifter 18 is
The difference between the current group scale factor and the block scale factor, ie, in this example, 2-1 = 1
Given as

[0083] Operation of the second shifter 18 is the same both in the second shifter 18, when the second shifter 18 of computing unit P ₀ will be described as an example, the second shifter 18 of computing unit P _0, the scale When "1" is given as a factor, the input arithmetic data Y (0) is shifted right by one bit by the right shifter 18a, and the upper 9 bits of the shifted arithmetic data Y (0) are shifted. Data (in this case, data of all bits) is output to the data rounding processing unit 18b. Then, the data rounding processing unit 18
b, the input 9-bit operation data Y (0)
, Add data "00000000"
"1" is added, and the upper 8 bits of the addition result are output to the product-sum operation unit 12. As a result, data rounding processing is performed on the operation data Y (0).
The arithmetic processing applied thereafter to the arithmetic data Y (0) to Y (3) is exactly the same as the arithmetic processing described for the above equation (1).

Next, the second data group including the operation data Y (4) to Y (7) is read out from the data memory 100, and the operation data Y (4) to Y (7) are transferred via the data bus 110, respectively. in is distributed to the second shifter 18 of each calculation unit P ₀ to P _3. The group scale factor for this data group, as shown in FIG.
It is "1". As a result, all the operation data of this data group is 1-1 = 0, that is, the second shifter 1
8 without being shifted.

At this stage, all operation data Y (0)
ＹY (7) is digit-aligned to the decimal point position of the maximum value of the data block. As a result, in the subsequent product-sum operation, fixed-point arithmetic can be performed. In this manner, in the present embodiment, the operation is performed on each data of the data group, the minimum scale factor representing the operation data is detected as the group scale factor, and each of the data is calculated based on the detected group scale factor. This series of processing for scaling the operation data is performed for each data group of the data block, and the smallest of the detected group scale factors is detected as a block scale factor. To perform the calculation again,
Before the operation, the second shifter 18 performs scaling on each operation data of the data group based on the group scale factor and the block scale factor of the data group, and rounds the scaled operation data. Processing is performed.

As a result, for the operation data belonging to the same data group, the data can be rounded at the same digit position. In addition, since the data can be rounded at a fixed digit position, the configuration is not so complicated. Therefore, compared to the related art, in the block floating method, it is possible to improve the calculation accuracy by rounding data with a relatively simple configuration.

In the above embodiment, the data memory 1
00 corresponds to the storage means according to claim 6 or 10,
The group scale factor register file 52 and the block scale factor register 56 are
The second shifter 18 corresponds to the first scaling means according to claims 6 to 8, 10, 12, or 13. In addition, the product-sum operation unit 12
The unit scale factor detector 14 corresponds to the calculating means according to claims 6, 7, 11 to 13.
Or corresponding to the scale factor calculation means described in 11,
The first shifter 16 corresponds to the second scaling means according to claim 6 or 12.

In the above embodiment, the group scale factor detector 50 corresponds to the group scale factor detecting means according to claim 6, and the block scale factor detector 54 corresponds to the block scale factor detector according to claim 6. Control logic 400 corresponding to the means;
Corresponds to the control means according to claim 6 or 10.

In the above embodiment, the second shifter 18 of each of the calculation units P _{0 to} P _{k -1} and the product-sum operation unit 12
Although the description has been made on the transfer of data between hardware and hardware through hardware means, the configuration shown in FIG. 6 can be specifically adopted. FIG. 6 is a block diagram showing a detailed configuration for transferring data between the second shifter 18 of each of the calculation units P _{0 to} P _k−1 and the product-sum operation unit 12.

In FIG. 6, the calculation unit P₀Is the second
A register 11a for holding data from the shifter 18 is provided.
I have. Calculation unit P₁, P_TwoIs the calculation unit
To P ₀, P₁Register holding data from the register 11a
The register 11a and the preceding calculation unit P₀, P₁Regis of
Data from the data 11a and data from the second shifter 18.
Selector that selects one of the
And a blade 11b. Calculation unit P_ThreeIs the first
Calculation unit P_TwoData from register 11a
And any of the data from the second shifter 18
The selector 11 b that outputs the result to the sum calculator 12 is provided.

Further, in the above-described embodiment, the unit scale factor detector 14 and the group scale factor detector 50 are used. . FIG.
FIG. 14 is a block diagram showing another configuration of the unit scale factor detector 14 and the group scale factor detector 50.

In FIG. 7, each of the calculation units P _{0 to} P
₄ is the unit scale factor detector 14,
It comprises a local block floating unit 15 composed of an XOR array 15a. And, it is configured to include a group scale factor detector 51 composed of the priority encoder 21 and a wired OR bus connecting the local block floating unit 15 of each of the calculation units P _{0 to} P ₄ and the group scale factor detector 51. I have.

Local block floating unit 15
Is configured to input operation data from the product-sum operation unit 12, perform an exclusive OR operation on adjacent bits of the operation data, and output the result. For example, when 4-bit operation data is input, the result of performing an exclusive OR operation on the fourth and third bits at the most significant position is defined as the third bit of the output data, and in the same manner,
The third and second operation results are the second bit of the output data, the second and first operation results are the first bit of the output data, and the most significant bit is the most significant bit of the output data. Output as

[0094] Thus, to calculate the scale factor of the calculated data from the multiply-add unit 12 of the computing unit P ₀ to P _4, even without calculating the group scale factor from their scale factor, each computing unit P ₀ It can be detected directly group scale factor from calculated data from the multiply-add unit 12 of the to P _4. Further, in the above embodiment, in order to facilitate understanding of the invention, the data group is configured to include data of the number corresponding to the number of the calculation units P _{0 to} P _{k−1. 0} to
Regardless of the number of P _k−1 , it may be configured to include a smaller number of data, or may be configured to include a larger number of data.

In the above embodiment, the control logic 400 is configured to control each unit according to the internal logic. However, the present invention is not limited to this.
Instead of 00, a data operation processing unit is configured by connecting a CPU, a RAM, and a ROM via a bus,
The program showing the above control procedure is executed by the CPU by R
You may comprise so that it may read from OM and execute. In this case, the program showing the above control procedure is stored in the ROM
However, the present invention is not limited to this, and the program may be read into a RAM from a storage medium storing a program indicating the above control procedure and executed.

Further, in the above embodiment, the data processing device according to the present invention is
Operation unit 200, control unit 300, control logic 400
And a hardware including an address generator 500. However, the present invention is not limited to this. For example, a control program for realizing the functions of the respective units by connecting a CPU, a RAM, and a ROM via a bus and using the CPU Some or all may be configured by software so that is read from the ROM and executed. Also in this case, the control program
The control program is stored in the OM, but is not limited thereto.
The program may be read into the AM and executed.

Here, the storage medium is a semiconductor storage medium such as a RAM or a ROM, a magnetic storage type storage medium such as an FD or HD, an optical read type storage medium such as a CD, CDV, LD, DVD, or the like. It is a magnetic storage type / optical read type storage medium, and includes any storage medium that can be read by a computer regardless of an electronic, magnetic, optical or other read method.

Further, in the above embodiment, the data arithmetic processing device and the data arithmetic processing program according to the present invention are applied to the case where the fixed-point arithmetic is performed by the block floating method as shown in FIG. The present invention is not limited to this, and can be applied to other cases without departing from the gist of the present invention.

[0099]

As described above, according to the data arithmetic processing apparatus according to the first to fifth aspects of the present invention, for the arithmetic data belonging to the same data group, the data is rounded at the same digit position. It can be performed. In addition, since the data can be rounded at a fixed digit position, the configuration is not so complicated. Therefore, in comparison with the related art, in the block floating method, it is possible to obtain an effect of improving the calculation accuracy by rounding data with a relatively simple configuration.

Further, claims 6 to 13 according to the present invention.
According to the described data operation processing device, for operation data belonging to the same data group, the data can be rounded at the same digit position. Further, since the first scaling means can perform data rounding processing at a fixed digit position, the configuration does not become so complicated. Therefore, in comparison with the related art, in the block floating method, it is possible to obtain an effect of improving the calculation accuracy by rounding data with a relatively simple configuration.

On the other hand, according to the data operation processing program of the present invention, the same effects as those of the data operation processing apparatus of the first aspect can be obtained. Further, according to the data operation processing program of the present invention, the same effect as the data operation processing device of the sixth aspect can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a data operation processing device according to the present invention.

FIG. 2 is a block diagram showing a configuration of a second shifter 18.

FIG. 3 is a diagram showing a product-sum operation procedure performed by each product-sum calculator 12.

FIG. 4 is a diagram for explaining a calculation process when the present invention is applied to a numerical example.

FIG. 5 is an example of a method of storing a data group and a group scale factor in association with each other.

FIG. 6 is a block diagram showing a detailed configuration for transferring data between the second shifter 18 of each of the calculation units P _{0 to} P _k−1 and the product-sum operation unit 12;

FIG. 7 is a block diagram showing another configuration of the unit scale factor detector 14 and the group scale factor detector 50.

FIG. 8 is a block diagram showing a configuration of a conventional data processing unit.

FIG. 9 is a diagram showing a data structure of data serving as an operation unit in the data memory 100;

[Explanation of symbols]

10, 18 Second shifter 12 Multiply-accumulate unit 14 Unit scale factor detector 16 First shifter 50, 51 Group scale factor detector 52 Group scale factor register file 54 Block scale factor detector 56 Block scale factor register 60 For digit alignment Adder 100 data memory 100 110 data bus 120 address bus 200 operation unit 300 control unit 400 control logic 500 address generator 110 data bus 120 data address bus 11a register 11b selector 15a XOR array 15 local block floating unit 21 priority encoder

Claims (15)

[Claims] 1. An apparatus for forming a data group including one or a plurality of data, further forming a data block including a plurality of data groups, and processing data in units of the data block, Calculate the scale factor of the data with the largest absolute value among the calculated data as the group scale factor.
These series of processes for scaling each of the operation data based on the detected group scale factor are performed for each data group of the data block, and further, the absolute value of the detected group scale factor is maximized. When the data corresponding to the operation data is detected as a block scale factor and the operation is performed again on the scaled operation data, before the operation, each operation data of the data group is subjected to the group of the data group. A data operation processing device, wherein scaling is performed based on a scale factor and the block scale factor, and data rounding is performed on the scaled operation data. 2. The method according to claim 1, wherein the operation is performed on data having a predetermined number of bits. For each operation data scaled based on the group scale factor and the block scale factor, From the operation data, data having a larger number of bits than the predetermined number of bits is extracted based on the most significant bit, and based on the value of the extracted data other than the data having the predetermined number of bits based on the most significant bit. A data arithmetic processing device for rounding the data. 3. The method according to claim 2, wherein, for each operation data scaled based on the group scale factor and the block scale factor, one bit of the operation data is smaller than the predetermined number of bits based on the most significant bit. It is characterized in that data of a large number of bits is extracted, and the data is rounded based on the value of data other than the data of the predetermined number of bits based on the most significant bit of the extracted data. Data processing unit. 4. The data rounding process according to claim 3, wherein in the data rounding process, one bit is added to a value of the extracted data other than the data having the predetermined number of bits based on the most significant bit, and the addition result is obtained. Wherein the data having the predetermined number of bits is output as an operation target based on the most significant bit of the data. 5. The method according to claim 1, wherein when performing the operation again on the scaled operation data, each operation data of the data group is processed before the operation. A data processing device, wherein scaling is performed based on a difference between a group scale factor and the block scale factor. 6. An apparatus comprising a plurality of calculation units, forming a data group including one or a plurality of data, further forming a data block including a plurality of data groups, and processing data in units of the data blocks Wherein each of the calculation units is a first scaling unit that performs scaling on data based on a given scale factor; an operation unit that performs an operation on data from the first scaling unit; A scale factor calculating unit that calculates a scale factor of the operation data from the operation unit; and a second scaling unit that performs scaling on the operation data from the operation unit based on another given scale factor, Further, storage means for storing data, and a scale factor for each of the calculation units A group scale factor detecting means for detecting, as the group scale factor, a scale factor corresponding to the operation data having the largest absolute value among the scale factors calculated by the calculating means; A block scale factor detecting means for detecting a value corresponding to the operation data having the maximum value as a block scale factor, and a control means for performing control for data processing, the control means comprising: Is read from the storage means, and for each data group of the data block, the data of the data group is distributed to the first scaling means of each of the calculation units, and the group scale detected by the group scale factor detecting means is read out. Given factor to the second scaling means of each calculation unit, the second of each calculation unit
When the operation data from the scaling unit is stored in the storage unit, and the operation is performed again on the operation data in the storage unit, the operation data of the data group is stored in each of the data groups of the data block. While allocating to the first scaling means of the calculation unit,
A scale factor comprising a difference between the group scale factor of the data group and the block scale factor is provided to first scaling means of each of the calculation units. And a data rounding process for the scaled data. 7. The data processing device according to claim 6, wherein the operation unit performs an operation on data of a predetermined number of bits, and wherein the first scaling unit performs the predetermined bit operation on the basis of the most significant bit from the scaled data. Data of a number of bits larger than the number of bits is extracted, and the data is rounded based on the value of data other than the data of the predetermined number of bits based on the most significant bit of the extracted data. A data processing unit characterized by the above. 8. The data processing device according to claim 7, wherein the first scaling means extracts, from the scaled data, data having a number of bits one bit larger than the predetermined number of bits with reference to the most significant bit, and extracts the extracted data. A data arithmetic processing device for performing a rounding process on the data based on a value of data other than the data of the predetermined number of bits based on the most significant bit of the data. 9. The data rounding process according to claim 8, wherein, in the data rounding process, one bit is added to a value of data other than the data having the predetermined number of bits based on the most significant bit of the extracted data, and the addition result is obtained. Wherein the data of the predetermined number of bits is output as a calculation target based on the most significant bit of the data. 10. The apparatus according to claim 6, further comprising: a second storage unit configured to store the group scale factor and the block scale factor, wherein the control unit controls the group scale factor and the block scale factor. Is stored in the second storage means in association with the calculation data, and when the calculation is performed again on the calculation data in the storage means, the corresponding group scale factor and block scale factor are read from the second storage means, For each data group of the data block, the operation data of the data group is distributed to the first scaling means of each of the calculation units, and a scale factor including a difference between the group scale factor of the data group and the block scale factor is calculated. Said A data processing unit, which is provided to a first scaling unit of each calculation unit. 11. The scale factor calculating means according to claim 6, wherein the scale factor calculating means calculates the number of redundant bits of the operation data from the operation means, and calculates this as a scale factor. A data processing device characterized by the above-mentioned. 12. The arithmetic unit according to claim 11, wherein the arithmetic unit is a fixed-point arithmetic unit, and the first scaling unit performs a bit shift on the data by a shift amount corresponding to a given scale factor. The data operation processing device according to claim 1, wherein said second scaling means performs a bit shift on operation data from said operation means by a shift amount corresponding to another given scale factor. 13. The arithmetic unit according to claim 6, wherein the arithmetic unit is a product-sum arithmetic unit that calculates a product sum of data from the first scaling unit and a predetermined coefficient. Data processing unit. 14. A computer-executable program for forming a data group including one or a plurality of data, forming a data block including a plurality of data groups, and processing data in units of the data block. Performing an operation on each data of the data group, and detecting a scale factor of the operation data having a maximum absolute value as a group scale factor;
These series of processes for scaling each of the operation data based on the detected group scale factor are performed for each data group of the data block, and further, the absolute value of the detected group scale factor is maximized. When the data corresponding to the operation data is detected as a block scale factor, and the operation is performed again on the scaled operation data, before the operation, each operation data of the data group is subjected to a group of the data group. A data operation processing program for performing scaling based on a scale factor and the block scale factor, and performing a data rounding process on the scaled operation data. 15. A computer having storage means for storing data, a data group including one or a plurality of data, and a data block including a plurality of data groups. A program for processing data in block units, wherein the first scaling means performs scaling on the data based on a given scale factor; the calculating means performs an operation on data from the first scaling means; Processing realized as a scale factor calculating means for calculating a scale factor of operation data from the operation means, and a second scaling means for scaling the operation data from the operation means based on another given scale factor Is executed for each data group. Group scale factor detecting means for detecting, as a group scale factor, a scale factor calculated by the scale factor calculating means corresponding to each of the data groups, the one corresponding to the operation data having the largest absolute value, and the group scale factor It is realized as a block scale factor detecting means for detecting, as a block scale factor, a data corresponding to the operation data having the largest absolute value among the group scale factors detected by the detecting means, and a control means for performing control for data processing. A program for executing a process, wherein the control unit reads data from the storage unit in units of the data block, and for each data group of the data block, stores the data of the data group in each of the data blocks. And assigns the group scale factor detected by the group scale factor detecting means to the second scaling means corresponding to each data group, and the second scaling means corresponding to each data group. When the operation data from the storage unit is stored in the storage unit, and the operation is performed again on the operation data in the storage unit, the operation data of the data group is stored in each of the data groups for each data group of the data block. And assigning a scale factor consisting of a difference between a group scale factor of the data group and the block scale factor to the first scaling means corresponding to each data group, A data calculation processing program for performing scaling on data based on a given scale factor, and performing data rounding processing on the scaled data.